hubertciebiada/context-crystallizer

**AI Context Engineering for Large Repositories**
Transform massive repositories into crystallized, AI-consumable knowledge through systematic analysis and optimization.
What is Crystallization?
Just as pressure transforms carbon into diamonds, Context Crystallizer applies systematic AI analysis to transform raw repositories into **crystallized knowledge** - structured, searchable, and optimized for AI consumption. Each file is analyzed to extract its purpose, key APIs, patterns, and relationships, creating a knowledge base that AI can efficiently search and understand.
Inspiration
Context Crystallizer was inspired by AI Distiller (aid), which pioneered the concept of intelligently extracting essential information from large codebases for AI consumption. While AI Distiller focuses on extracting public APIs and type information using tree-sitter parsers, Context Crystallizer takes a complementary approach by having AI agents generate comprehensive crystallized contexts about functionality, patterns, and relationships.
The Problem 🔥
AI agents hit context length limits when working with large repositories. A typical enterprise repository has 10,000+ files, but LLMs can only process a fraction at once. This forces AI to work blindly or make assumptions about unfamiliar code.
The Solution ✨
Context Crystallizer creates a **searchable crystallized context base** of AI-optimized knowledge:
- 🔍 **Search by functionality**: "find authentication middleware"
- ⚡ **Token-efficient**: 5:1 compression ratio (source:crystallized context)
- 🤖 **AI-optimized format**: Structured specifically for LLM consumption
- 📊 **Smart assembly**: Combines multiple contexts within token limits
- 💎 **Crystallized knowledge**: Preserves essential information in optimized form
How It Works
**Simple 3-step crystallization process:**
1. **Initialize**: Scan repository and prepare for crystallization
- Automatically respects .gitignore patterns
- Skips common build directories (node_modules, dist, build, .git)
- Filters out binary files and very large files (>1MB)
2. **Crystallize**: AI analyzes each file to extract meaningful knowledge
3. **Search**: Find relevant crystallized contexts for any task
Quick Demo
**Developer using Claude Code with a large documentation repository:**
> **Developer**: "I need to understand how authentication works in this massive project"
>
> **Claude**: "I'll crystallize this repository first to build a searchable knowledge base, then find all authentication-related information."
>
> *Claude crystallizes the repository - scanning and analyzing each file*
>
> **Claude**: "Crystallization complete! I found 5 files with authentication logic. The main JWT middleware handles token validation with Redis session caching. Here's how it works..."
>
> *Claude provides comprehensive explanation using crystallized contexts*
>
> **Developer**: "What files depend on the authentication system?"
>
> **Claude**: "Let me search for related crystallized contexts..."
>
> *Claude uses find_related_crystallized_contexts() to discover dependencies*
Installation & Setup
Quick Start (5 minutes)
Claude Desktop Integration
Add to your Claude Desktop configuration (~/claude_desktop_config.json):
Context Crystallizer Tools
Context Crystallizer provides dual access to crystallization functionality: AI agents can use MCP tools for conversation-driven analysis, while developers can use CLI commands for direct control.
AI Agent Usage (MCP Tools)
AI agents interact with Context Crystallizer through MCP (Model Context Protocol) for conversation-driven crystallization and knowledge search.
| Tool | Purpose | AI Agent Conversation Example |
|------|---------|-------------------------------|
| **get_crystallization_guidance** | Get comprehensive analysis guidance | **Developer**: "How should I analyze these files?" **Claude**: "Let me get the crystallization guidance..." *Claude calls get_crystallization_guidance* **Claude**: "Here's the complete analysis methodology with templates, quality standards, and AI-specific guidance for creating crystallized contexts." |